Study On Filtering And Building-footprint Segmentation Of Airborne Lidar Data

As a new integrated technology, airborne light detection and ranging (LIDAR) technology are widely used in geographical conditions monitoring, topographic survey and3D city modeling, etc. Point cloud filtering and building-footprint segmentation are discussed and researched in this thesis.In order to manage the massive amounts of airborne LIDAR point cloud, virtual grid will be used in this thesis. Virtual grid can retain the original information of point cloud, at the same time it makes data organization and neighborhood selection convenient and efficient. After studding on the characteristics of point cloud and the existing filtering algorithm, this thesis tries to improve mathematical morphological filtering algorithms. Compared to the traditional, there are two improvements in the improved algorithm:ameliorate the fixed filtering window size and elevation difference threshold to gradient window size and variable elevation difference threshold.Filtering method based on morphological gradient and regional growth is described in detail. This method takes morphological gradient of each point as the prerequisite. If morphological gradient of one point satisfied a certain condition, then mark the point cloud. Do morphological open operation on the points include in the neighborhood of the marked point. Separating the ground and non-ground points is based on the elevation difference before and after the open operation. Repeat the above steps, until meet the stopping criteria. In order to avoid the excessive filtering, region growth is added and used. It can improve the accuracy of the filtering effectively. Experiment show that, filtering method based on morphological gradient and region growth can effectively improved the filtering effect at the edges of the building and the tall vegetation zone.Building model can be seen as some plane, this thesis use the improved random Hough transform(RHT) method spilt the point cloud that contain in roof plane of the building. The improved RHT method uses a pure mathematical principle in order to detect the best planes from a point cloud. The RHT requires the use of four parameters:the step onθ, φ and ρ axis and the threshold entering in the region growing algorithm. In this thesis, the improved RHT use two conditions—the distance between point to plane and the distance and elevation difference between two points—to control plane growth automatically. Those conditions can effectively improve the efficiency and accuracy of building-footprint segmentation algorithm.At the end, the use of C#.NET and AutoCAD.NET technology has carried on the secondary development on AutoCAD, and write the filtering and building-footprint segmentation of airborne LIDAR data program—LIDAR-DMP. The program can be achieved several functions: read in a variety of formats data, show and render the point cloud, organize the point cloud data, filtering of the LIDAR data, segmentation the building-footprint from the nonground points and so on.